What problem does it solve?
Coordinating multi-agent execution of a complex requirement is hard: tasks have dependencies, agents duplicate codebase exploration, and results from earlier work never reach later tasks. This Skill turns a plain-language requirement into a dependency-sorted CSV task plan and runs it wave by wave with cross-wave context propagation.
Core Features & Use Cases
- Requirement Decomposition: Uses a CLI analysis pass to break a requirement into 3-10 atomic tasks with test cases, acceptance criteria, scopes, and dependencies, then computes execution waves via topological sort.
- Wave-Based Concurrent Execution: Runs each wave through spawn_agents_on_csv with configurable concurrency, injecting previous-wave findings into a prev_context column so downstream tasks inherit upstream results.
- Shared Discovery Board: All agents append code patterns, integration points, and blockers to a shared discoveries.ndjson file, eliminating redundant exploration across waves.
- Use Case: Ask it to "Implement user authentication with OAuth, JWT, and 2FA" — it plans the tasks, runs independent setup work concurrently in Wave 1, then executes dependent integration tasks in later waves with full context from earlier findings.
Quick Start
Run the csv-wave-pipeline skill with a requirement like "Implement user authentication with OAuth, JWT, and 2FA" and approve the generated task breakdown to start wave execution.
